This example is adapted from Soon-Yau Cheong's excellent book "Hands-On Image Generation with TensorFlow": https://www.amazon.com/dp/1838826785 -- check it out!
Show this thread
New example on http://keras.io : generate faces that don't exist with this complete StyleGAN implementation -- in less than 500 lines of code. https://keras.io/examples/generative/stylegan/ …pic.twitter.com/wHfkOa0Eqs
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Hello. In your opinion, when training from scratch, what is the minimum size of a dataset to train styleGAN ? (I work on monkey faces, in behavioral ecology, so transfer learning is not necessarily adapted)
-
You dense sampling of the latent manifold (so can do interpolation on the manifold). What "dense" means depends on the variety/complexity of your data. Monkey faces all precisely facing the camera -- I'd guess you need 5-10k images for decent results and 50k for great results.
- Show replies
New conversation -
-
-
it doesn't even look like human face
-
The example uses low resolution images to make it easier for people to try. Check out my other notebook that gives better image fidelity https://github.com/PacktPublishing/Hands-On-Image-Generation-with-TensorFlow-2.0/blob/master/Chapter07/ch7_faster_stylegan.ipynb …pic.twitter.com/KyywaZ9dzV
End of conversation
New conversation -
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Compared to pytorch, we don't get reproducible results in keras / tensorflow. Which is why tensorflow is hated by researchers.. Run twice and you random initialization you do..
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I find it very hard to get the celeb_a dataset using tfds.load(). Can you please add a mechanism to feed the data that is already downloaded in a folder?
-
I tried a workaround using image_dataset_from_directory(). But for some reason, it always throws an error when reaching the end of batch. I guess because image_dataset_from_directory does not have drop_remainder?
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.